Intelligent and reconfigurable ultra-wideband spectrum characterization at sub-nyquist rate

Abstract

Historically, throughput is one of the key performance indicators driving the transition to next-generation cellular networks. The throughput per square kilometer depends on three factors: 1) Available spectrum, 2) Base station density, and 3) Spectrum utilization efficiency. The mmWave spectrum (24 GHz - 100GHz) is actively being explored to augment the sub-6 GHz spectrum (450 MHz6000 MHz) due to the availability of a wide spectrum and low auction cost. However, it has limited coverage and range, limiting its usefulness in indoor short-range mobile broadband services. This makes the sub-6 GHz spectrum a preferred candidate for outdoor communications and network coverage services. The high auction cost of the sub-6 GHz spectrum limits the licensed spectrum, and base station density is constrained due to infrastructure cost, handover overhead, and interference constraints. Thus, innovative ways to utilize the sub-6 GHz spectrum efficiently needs to be explored. One promising solution is dynamic spectrum sharing which is now a de facto approach in cellular networks. For instance, 5G supports the deployment in shared (2.3 GHz Europe / 3.5 GHz USA) and unlicensed (2.4 GHz / 5-7 GHz / 57-71 GHz global) spectrums along with licensed noncontiguous spectrum. Joint radar-communication systems are being explored to improve the utilization of a large section of the sub-6 GHz spectrum allocated to radar applications. Similarly, IEEE 802.15.4 for industrial internet-of-things (IIoT) networks support deployment in 250-740 MHz, 3.1 4.8 GHz and 6 - 11.6 GHz. To enhance spectrum efficiency, multi-antenna systems are being explored, allowing multiple users to communicate simultaneously over a given frequency band. This demands wideband spectrum analyzer (WSA) for the digitization of ultra-wide non-contiguous spectrum (UWNS), and capability to identify the transmission opportunities in time, frequency and spatial domains reliably. The traditional approaches need complex hardware and signal processing algorithms that question their suitability for real-time requirements. iii In this thesis, we focus on the sub-Nyquist sampling (SNS) and sparse antenna array based intelligent and reconfigurable WSA for the digitization and spatial sensing of UWNS using low-rate analog-to-digital converters (ADCs). In the first contribution, we explore reconfigurable SNS, which allows the digitization of a non-contiguous spectrum. The non-contiguous nature demands learning the occupancy of various parts of the spectrum since spectrum digitization can fail when the number of occupied bands in a digitized spectrum is higher than that of ADCs. On the other hand, high throughput requirement demands digitization of as wide spectrum as possible. We address such a trade-off via Multi-Play Multi-Armed Bandit (MPMAB) framework. The functionality of the proposed intelligent and reconfigurable WSA is validated using real radio signals via universal software radio peripheral (USRP) testbed. After successful digitization and identification of vacant spectrum, the next contribution deals with the characterization of the occupied spectrum. We extend the WSA using a multi-antenna approach to enable blind identification of carrier frequency, angle of arrival and modulation scheme. It is referred to as ultra-wideband angular spectrum sensing (UWASS). The UWASS receiver overcomes the limitation of existing methods in which the number of antennas depends on the spectrum sparsity making it computationally efficient. The performance of the UWASS receiver is analyzed for uniform and sparse antenna arrays. In the third contribution, we develop a realistic multi-antenna USRP testbed to demonstrate the functional correctness of the UWASS receiver for various parameters such as signal-to noise ratio (SNR), spectrum sparsity, antenna array, and its size. Recently, deep learning has outperformed conventional statistical and machine learning based spectrum characterization methods. In the fourth and last contribution, we explored various deep learning approaches for spectrum reconstruction and characterization. Specifically, we propose a novel non iterative wideband deep learning-based modulation classification (WDLMC) which can simultaneously identify the frequency band status and the modulation scheme of all the frequency bands in the digitized spectrum compared to existing iterative approaches. We also propose deep learning based spectrum reconstruction for UWASS as an alternative to the conventional orthogonal matching pursuit (OMP) approach. In-depth performance analysis validates the functional correctness and superiority of the proposed approach over state-of-the-art approaches in terms of computational complexity and execution time. iv to summarize, the proposed intelligent and reconfigurable WSA offer efficient and hardware friendly solutions to improve the utilization of the sub-6 GHz spectrum by identifying the spectrum opportunities in time, frequency and spatial domains.